How the weights of different loss functions affect performance of the network in GAN based SISR?

[KhushbooChauddhary]

Total perceptual loss in SRGAN paper is weighted sum of content loss and adversarial loss.

Total loss = Content loss + (10^(-3)) Adversarial loss Please tell why 10^(-3) is used? what is its impact on performance if some other value is used ?? or does it affects number of iterations for training of network ?

[hanjr92]

sorry too late to reply you. In paper, we can find the total loss = content loss + (10^(-3)) Adversarial loss. @KhushbooChauddhary . I think this value is a hyperparameter related to the dataset. You need to debug on different datasets.